VideoTask | NeMo Curator

VideoTask is the task type for video processing in NeMo Curator.

Import

from nemo_curator.tasks import VideoTask

Class Definition

from dataclasses import dataclass
from nemo_curator.tasks.video import Video
@dataclass
class VideoTask(Task[Video]):
    """Task containing a single video for processing.
    Attributes:
        task_id: Framework-managed deterministic lineage identifier.
        dataset_name: Name of the source dataset.
        data: Video object with path and metadata.
    """
    dataset_name: str
    data: Video
    # task_id is inherited from Task with init=False.

Video Class

The video data is represented by a Video object:

@dataclass
class Video:
    """Represents a video file with metadata.
    Attributes:
        path: Path to the video file.
        start_time: Start time in seconds (for clips).
        end_time: End time in seconds (for clips).
        metadata: Additional metadata dictionary.
        embeddings: Optional embedding vector.
    """
    path: str
    start_time: float | None = None
    end_time: float | None = None
    metadata: dict[str, Any] = field(default_factory=dict)
    embeddings: np.ndarray | None = None

Properties

`num_items`

Get the number of items (always 1 for VideoTask).

@property
def num_items(self) -> int:
    """Returns 1 (VideoTask represents a single video)."""

Creating VideoTask

from nemo_curator.tasks import VideoTask
from nemo_curator.tasks.video import Video
# Create a video object
video = Video(
    path="/data/videos/video1.mp4",
    start_time=0.0,
    end_time=30.0,
    metadata={
        "duration": 120.5,
        "fps": 30,
        "resolution": "1920x1080",
    },
)
# Create task
task = VideoTask(
    dataset_name="video_dataset",
    data=video,
)

Do not pass task_id to the constructor. The framework assigns it after the task crosses a stage boundary.

Usage in Stages

from dataclasses import dataclass
from nemo_curator.stages.base import ProcessingStage
from nemo_curator.tasks import VideoTask
@dataclass
class VideoFilterStage(ProcessingStage[VideoTask, VideoTask]):
    """Filter videos based on duration."""
    name: str = "VideoFilter"
    min_duration: float = 5.0
    max_duration: float = 300.0
    def inputs(self) -> tuple[list[str], list[str]]:
        return ["data"], []
    def outputs(self) -> tuple[list[str], list[str]]:
        return ["data"], []
    def process(self, task: VideoTask) -> VideoTask | None:
        video = task.data
        duration = video.metadata.get("duration", 0)
        if not (self.min_duration <= duration <= self.max_duration):
            return None
        return VideoTask(
            dataset_name=task.dataset_name,
            data=video,
            _metadata=task._metadata,
            _stage_perf=task._stage_perf,
        )

Common Operations

Splitting Videos into Clips

The video splitting stages return multiple VideoTask objects:

def process(self, task: VideoTask) -> list[VideoTask]:
    clips = []
    video = task.data
    for i, (start, end) in enumerate(self._compute_splits(video)):
        clip_video = Video(
            path=video.path,
            start_time=start,
            end_time=end,
            metadata=video.metadata.copy(),
        )
        clips.append(VideoTask(
            dataset_name=task.dataset_name,
            data=clip_video,
            _metadata=task._metadata,
            _stage_perf=task._stage_perf,
        ))
    return clips

Adding Embeddings

def process(self, task: VideoTask) -> VideoTask:
    video = task.data
    video.embeddings = self.encoder.encode(video.path)
    return VideoTask(
        dataset_name=task.dataset_name,
        data=video,
        _metadata=task._metadata,
        _stage_perf=task._stage_perf,
    )

Source Code

View source on GitHub